[Issue 15064] [CTFE] AliasSeq in multi-level alias this fails in CTFE

[d-bugmail at puremagic.com]

https://issues.dlang.org/show_bug.cgi?id=15064

Simen Kjaeraas <simen.kjaras at gmail.com> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
                 CC|                            |simen.kjaras at gmail.com
          Component|phobos                      |dmd
            Summary|[CTFE] std.range.enumerate  |[CTFE] AliasSeq in
                   |is not CTFE-able            |multi-level alias this
                   |                            |fails in CTFE
           Severity|enhancement                 |normal

--- Comment #5 from Simen Kjaeraas <simen.kjaras at gmail.com> ---
Reduced test case:

import std.meta : AliasSeq;

struct Super {
    AliasSeq!(int, int) expand;
    alias expand this;
}

struct Tuple {
    Super inner;
    alias inner this;
}

struct Range {
    Tuple front;
    void popFront() {}
    bool empty = true;
}

int test1() {
    import std.algorithm : map;
    foreach (i, e; Range.init) {} // Fails
    foreach (i, e; [Tuple.init].map!(a=>a)) {} // Fails
    foreach (i, e; [Tuple.init]) {} // Works [line 23]
    return 3;
}

enum s1 = test1();

And even more reduced, I think:

struct Super2 {
    AliasSeq!int expand;
    alias expand this;
}

struct Tuple2 {
    Super2 inner;
    alias inner this;
}

int test2() {
    AliasSeq!int a = Tuple2.init; // Fails with same message
    int b = Tuple2.init[0]; // Works [line 41]

    return 3;
}

enum s2 = test2();

Since test2 gives the same error message, and the same kind of unpacking will
happen in both cases, I assume that's the root problem.

So in summary: If an AliasSeq is used in alias this inside another alias this,
some context goes missing in CTFE, possibly only during tuple assignment (see
line 41). Strangely, it seems to work when the unpacking comes from an array
instead of a range (line 23). Not sure why that is, but it seems tangential to
the real problem here.

--